In [11]:
"""
Created on Mon May 21 10:24:12 2018
@author: XieMN
"""
from PIL import Image
import matplotlib.pyplot as plt
import math
import copy
%matplotlib inline
class BilateralFilter(object):
"""
定义双边滤波器类：通过空间域和值域模板进行平滑
变量：
ds：distance sigma
rs: range sigma
c_weight_table：空间域权重模板
s_weight_table: 值域权重模板
radius: 平滑模板的半径，即平滑模板大小为（2*radius+1）
"""
def __init__(self,ds, rs, radiu):
"""初始化变量"""
self.ds = ds
self.rs = rs
self.c_weight_table = []
self.s_weight_table = []
self.radius = radiu
def build_distance_weight_table(self):
"""定义空间域模板"""
# size = 2 * self.radius + 1
for semi_row in range(-self.radius,self.radius+1):
self.c_weight_table.append([])
for semi_col in range(-self.radius,self.radius+1):
# calculate Euclidean distance between center point and close pixels
delta = -(semi_row * semi_row + semi_col * semi_col)/(2*(self.ds**2))
self.c_weight_table[semi_row+self.radius].append(
math.exp(delta))
def build_similarity_weight_table(self):
"""定义值域模板"""
for i in range(256): # since the color scope is 0 ~ 255
delta = -(i * i) / (2*(self.rs**2))
self.s_weight_table.append(math.exp(delta ))
def clamp(self,p):
"""return RGB color between 0 and 255"""
if p < 0:
return 0
elif p > 255:
return 255
else:
return p
def bilateral_filter(self, src):
"""
双边滤波器的方法：对原始图进行双边滤波器平滑，并返回平滑过后的图
"""
height = src.size[0]
width = src.size[1]
radius = self.radius
self.build_distance_weight_table()
self.build_similarity_weight_table()
in_pixels = src.load()
raw_data=[]
out_data=copy.deepcopy(src)
# out_pixels = {}
red_sum = green_sum = blue_sum = 0 # result of convolution before normalization
cs_sum_red_weight = cs_sum_green_weight = cs_sum_blue_weight = 0 # normalization
#对于边缘像素采用不平滑的方法
for row in range(radius,height-radius):
for col in range(radius,width-radius):
# 对每个像素进行平滑
tr = in_pixels[row, col][0]
tg = in_pixels[row, col][1]
tb = in_pixels[row, col][2]
raw_data.append((tr, tg, tb))
for semi_row in range(-radius, radius+1):
for semi_col in range(-radius, radius+1):
# 获得模板内的像素
row_offset = row + semi_row
col_offset = col + semi_col
tr2 = in_pixels[row_offset, col_offset][0]
tg2 = in_pixels[row_offset, col_offset][1]
tb2 = in_pixels[row_offset, col_offset][2]
#卷积计算
cs_red_weight = (
self.c_weight_table[semi_row+radius][semi_col+radius]
* self.s_weight_table[(abs(tr2 - tr))]
)
cs_green_weight = (
self.c_weight_table[semi_row+radius][semi_col+radius]
* self.s_weight_table[(abs(tg2 - tg))]
)
cs_blue_weight = (
self.c_weight_table[semi_row+radius][semi_col+radius]
* self.s_weight_table[(abs(tb2 - tb))]
)
cs_sum_red_weight += cs_red_weight
cs_sum_blue_weight += cs_blue_weight
cs_sum_green_weight += cs_green_weight
red_sum += cs_red_weight * float(tr2)
green_sum += cs_green_weight * float(tg2)
blue_sum += cs_blue_weight * float(tb2)
# 归一化过程
tr = int(math.floor(red_sum / cs_sum_red_weight))
tg = int(math.floor(green_sum / cs_sum_green_weight))
tb = int(math.floor(blue_sum / cs_sum_blue_weight))
temp_rgb=(self.clamp(tr), self.clamp(tg), self.clamp(tb))
out_data.putpixel((row,col),temp_rgb)
# clean value for next time
red_sum = green_sum = blue_sum = 0
cs_red_weight = cs_green_weight = cs_blue_weight = 0
cs_sum_red_weight =cs_sum_blue_weight =cs_sum_green_weight = 0
return out_data
做不同sigma比较¶
In [12]:
def main():
# DS = 1.0 #defualt distance sigma
# DR = 30.0 #defualt range sigma
# #RADIUS = int(max(DS,DR)) #defualt radius
# RADIUS = 3
# bf = BilateralFilter(DS,DR,RADIUS)
img0 = Image.open('lena_10.jpg')
bf = BilateralFilter(1,25,3)
dest0 = bf.bilateral_filter(img0)
dest0.save('lena_10_m_3_s_1_c_25.jpg')
bf = BilateralFilter(25,25,3)
dest1 = bf.bilateral_filter(img0)
dest1.save('lena_10_m_3_s_25_c_25.jpg')
bf = BilateralFilter(25,1,3)
dest2 = bf.bilateral_filter(img0)
dest2.save('lena_10_m_3_s_25_c_1.jpg')
plt.figure(figsize = (18,12))
plt.subplot(2,2,1)
plt.title("Origin",fontsize=20)
plt.imshow(img0)
plt.axis('off')
plt.subplot(2,2,2)
plt.title("lena_10_m_3_s_1_c_25",fontsize=20)
plt.imshow(dest0)
plt.axis('off')
plt.subplot(2,2,3)
plt.title("lena_10_m_3_s_25_c_25",fontsize=20)
plt.imshow(dest1)
plt.axis('off')
plt.subplot(2,2,4)
plt.title("lena_10_m_3_s_25_c_1",fontsize=20)
plt.imshow(dest2)
plt.axis('off')
plt.show()
if __name__ == '__main__':
main()
验证sigma取值越大，图像越模糊，>150时基本为卡通图像¶
In [14]:
def main():
# DS = 1.0 #defualt distance sigma
# DR = 30.0 #defualt range sigma
# #RADIUS = int(max(DS,DR)) #defualt radius
# RADIUS = 3
# bf = BilateralFilter(DS,DR,RADIUS)
img0 = Image.open('lena_10.jpg')
bf = BilateralFilter(10,10,3)
dest0 = bf.bilateral_filter(img0)
dest0.save('lena_10_m_3_s_10_c_10.jpg')
bf = BilateralFilter(60,60,3)
dest1 = bf.bilateral_filter(img0)
dest1.save('lena_10_m_3_s_60_c_60.jpg')
bf = BilateralFilter(150,150,3)
dest2 = bf.bilateral_filter(img0)
dest2.save('lena_10_m_3_s_150_c_150.jpg')
plt.figure(figsize = (18,12))
plt.subplot(2,2,1)
plt.title("Origin",fontsize=20)
plt.imshow(img0)
plt.axis('off')
plt.subplot(2,2,2)
plt.title("lena_10_m_3_s_10_c_10",fontsize=20)
plt.imshow(dest0)
plt.axis('off')
plt.subplot(2,2,3)
plt.title("lena_10_m_3_s_60_c_60",fontsize=20)
plt.imshow(dest1)
plt.axis('off')
plt.subplot(2,2,4)
plt.title("lena_10_m_3_s_150_c_150",fontsize=20)
plt.imshow(dest2)
plt.axis('off')
plt.show()
if __name__ == '__main__':
main()
做不同模板大小m比较¶
In [17]:
def main():
# DS = 1.0 #defualt distance sigma
# DR = 30.0 #defualt range sigma
# #RADIUS = int(max(DS,DR)) #defualt radius
# RADIUS = 3
# bf = BilateralFilter(DS,DR,RADIUS)
img0 = Image.open('lena_10.jpg')
bf = BilateralFilter(10,25,3)
dest0 = bf.bilateral_filter(img0)
dest0.save('lena_10_m_9_s_10_c_25.jpg')
bf = BilateralFilter(10,25,5)
dest1 = bf.bilateral_filter(img0)
dest1.save('lena_10_m_11_s_10_c_25.jpg')
bf = BilateralFilter(10,25,7)
dest2 = bf.bilateral_filter(img0)
dest2.save('lena_10_m_13_s_10_c_25.jpg')
plt.figure(figsize = (18,12))
plt.subplot(2,2,1)
plt.title("lena_10_Origin",fontsize=20)
plt.imshow(img0)
plt.axis('off')
plt.subplot(2,2,2)
plt.title("lena_10_m_9_s_10_c_25",fontsize=20)
plt.imshow(dest0)
plt.axis('off')
plt.subplot(2,2,3)
plt.title("lena_10_m_11_s_10_c_25",fontsize=20)
plt.imshow(dest1)
plt.axis('off')
plt.subplot(2,2,4)
plt.title("lena_10_m_13_s_10_c_25",fontsize=20)
plt.imshow(dest2)
plt.axis('off')
plt.show()
if __name__ == '__main__':
main()
做不同高斯噪声下的图，DS和DR最佳的选择实验¶
In [16]:
def main():
# DS = 1.0 #defualt distance sigma
# DR = 30.0 #defualt range sigma
# #RADIUS = int(max(DS,DR)) #defualt radius
# RADIUS = 3
# bf = BilateralFilter(DS,DR,RADIUS)
img0 = Image.open('lena_10.jpg')
img1 = Image.open('lena_20.jpg')
bf = BilateralFilter(10,25,5)
dest0 = bf.bilateral_filter(img0)
dest0.save('lena_10_m_5_s_10_c_25.jpg')
bf = BilateralFilter(10,50,5)
dest1 = bf.bilateral_filter(img1)
dest1.save('lena_10_m_5_s_10_c_50.jpg')
plt.figure(figsize = (18,12))
plt.subplot(2,2,1)
plt.title("lena_10_Origin",fontsize=20)
plt.imshow(img0)
plt.axis('off')
plt.subplot(2,2,2)
plt.title("lena_10_m_5_s_10_c_25",fontsize=20)
plt.imshow(dest0)
plt.axis('off')
plt.subplot(2,2,3)
plt.title("lena_20_origin",fontsize=20)
plt.imshow(img1)
plt.axis('off')
plt.subplot(2,2,4)
plt.title("lena_20_m_5_s_10_c_50",fontsize=20)
plt.imshow(dest1)
plt.axis('off')
plt.show()
if __name__ == '__main__':
main()
